The CFG2_cross_matching.md notebook is still under review and is not fully supported yet. Once it passes the review, it will have its own environment (kernel) like the the other notebooks.
As already mentioned, this notebook is not yet ready, but thanks for taking a stab at running it. I am on the small default astro server right now and have ran through it without issue (thought I think the requirements.txt has been changed following your description above, I am running the version in shared-storage/heasarc-tutorials).
You touch on an important point though. Cell 30 is the bulk query, sending 340 Fermi sources to the GAIA and 2MASS TAP services and right now is not reliable because it performs a single query on each Fermi source. The best (fastest, easiest, all the things) is to provide an upload table in the TAP query, which is going to be offered in the final version (or with astroquery, I am testing which one will be the best to offer).